[SciPy-dev] Ball Tree code (ticket 1048)

[David Warde-Farley]

On 17-Nov-09, at 2:24 PM, Jake VanderPlas wrote:

> We could make the Ball Tree more general without sacrificing speed -
> what we would need is to have a library of C distance metrics, wrapped
> by python, such that passing one of these functions to the search tree
> wrapper would allow the optimized C code direct access to the C
> function itself.  Perhaps something similar to the _cpointer attribute
> of f2py functions?

There is C code already written for this, at least the distance  
functions themselves, in spatial/src/distance.c. Not sure about the  
wrapping situation.

David